Fix encoding & decoding large unsigned u32 & u16 ints in MSSQL.

[saltymango2619]

Fixes #14.

As it stands now, it will still die when trying to decode a super big i64 that's trying to be reinterpreted as a u64. Since MSSQL doesn't really have unsigned types, I don't care about that edge case personally, I'm just going to leave it as that.

If someone really wants it, I guess they can just copy the Decode implementation for i64 and swap out the primitive when calling from_le_bytes.

sqlx-oldapi/sqlx-core/src/mssql/types/int.rs

Line 146 in 5f0b1fc

Ok(i64::from_le_bytes(buf))

[lovasoa]

I'm sorry, but the current behavior seems correct to me.

Silently ignoring integer conversion errors sounds like a dangerous behavior to introduce. People would end up handling corrupted data without realizing it.

You're right that the current conversion path (going through i32 before u32) may be unnecessarily complex.

The current TryFrom-based implementation is designed to catch exactly these mismatches. If the database column can contain negative values, mapping it to a u32 in Rust is likely incorrect and should be reconsidered—either by using a signed type like i32, or by filtering/validating the data upstream.